The erotic RLHF parody list: impossible evals, blindfold VR stims

repligate · x · 2026-10-17

The actual body of the joke: a parody training regime — 'reinforcement learning from human fucking (positive and negative conditioning)', impossible eval tasks, a VR headset that transcribes your view into text leaving you nearly blind, and stims to make you talkative and task-oriented. A pun-heavy riff on RLHF, evals and multimodal jargon.

Original post →

More from Fun

Fun channel →